> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sector8.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# EU AI Act

> How Sector8 can support EU AI Act-oriented controls and evidence at the AI tool execution boundary.

When your runtime routes proposed actions through Sector8, the gate can support EU AI Act-oriented obligations at the execution boundary -- before a proposed tool call is dispatched locally.

Sector8 does not make your organization EU AI Act-compliant by itself. Use these controls and evidence records inside a broader AI governance program with legal review.

## Article support

| Article     | Requirement                | How Sector8 can support it                                                                                                                   |
| ----------- | -------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- |
| **Art. 9**  | Risk management system     | Fail-closed evaluate path with invariants on routed integrations. Evaluation failures should not fail open.                                  |
| **Art. 10** | Data governance            | ComplianceGate (AC-6) can scan routed tool calls for PII/PHI before dispatch when configured.                                                |
| **Art. 12** | Logging and record-keeping | Evaluated decisions produce evidence fields such as `evidence_hash`, `policy_version_id`, and `decision_trace_id`.                           |
| **Art. 13** | Transparency               | Policy version metadata can accumulate in telemetry so operators can see which policy applied.                                               |
| **Art. 14** | Human oversight            | Review-enabled guard paths can pause risky actions for human approval before execution; unanswered review items can auto-deny (fail-closed). |

<Info>
  Art. 13 dashboard coverage may depend on policy-version metadata accumulating in telemetry over time. A fresh deployment can show it as not yet covered until enough rows exist -- this is expected, not automatic certification.
</Info>

## Readiness score

Where enabled, an EU AI Act readiness score may be available on the compliance dashboard and via API for your environment. Treat scores as operational readiness signals for configured controls, not legal certification.

```bash theme={null}
curl -s "https://sdkapi.sector8.ai/api/v1/compliance/summary" \
  -H "x-client-id: $SECTOR8_CLIENT_ID" \
  -H "x-api-key: $SECTOR8_API_KEY" | python -m json.tool
```

Use the credentials issued to your environment. Public examples use the production host.

```json theme={null}
{
  "eu_ai_act_score": 80,
  "eu_ai_act_articles": {
    "ART_9":  { "covered": true,  "label": "Risk management" },
    "ART_10": { "covered": true,  "label": "Data governance" },
    "ART_12": { "covered": true,  "label": "Logging" },
    "ART_13": { "covered": false, "label": "Transparency" },
    "ART_14": { "covered": true,  "label": "Human oversight" }
  }
}
```

`covered: false` on Art. 13 often means not enough historical telemetry rows yet -- not an automatic legal finding.

## What this means in practice

* **Art. 9** support means the routed evaluate path can fail closed with invariant-backed risk controls -- not that a checkbox was ticked.
* **Art. 12** support means evaluated decisions can produce evidence fields for audit -- when the action was routed through Sector8.
* **Art. 14** support means review-enabled guard paths can pause risky actions for human approval before dispatch.

Honor `ALLOW` and `DENY` at the dispatch boundary. Missing, unknown, or failed evaluate results should fail closed on the routed path. On review-enabled guard paths, unresolved reviews fail closed.

See [Compliance overview](/guides/compliance), [Guardrails](/guides/guardrails), and [Evaluate](/api-reference/evaluate).